Re: st: large sample bias

[Austin Nichols <austinnichols@gmail.com>]

Bülent Köksal <bkoksal@gmail.com>:
Larger samples are always better. That said, if your sample is getting
bigger because you adding observations between other observations, you
have "infill" asymptotics and should worry even more about measurement
error.  In panel models, measurement error can be even more of a
problem (than for pooled OLS) since you are typically throwing out
some or all of the between variation that is less affected by
measurement error--if your sample size gets larger by exploiting
higher and higher frequency data, you may be adding more and more
measurement error --but you would need validation data to know whether
this a real problem or not.

On Thu, Jan 3, 2013 at 10:17 AM, Bülent Köksal <bkoksal@gmail.com> wrote:
> Dear All,
>
> Is there a concept called large sample bias? If so, can this occur for
> a panel fixed effects estimation?
>
> When we have a sufficiently large sample (as in some market
> microstructure studies that use high-frequency data), probably all RHS
> variables will be significant. This is expected as the large sample
> size allows us to estimate the parameters with high precision. Now I
> understand that when the sample size is huge, it would be a good idea
> to include some sort of discussion about economic significance of the
> coefficients since these coefficients will be statistically
> significant even if they are very close to zero.
>
> But are there any other problems with a large sample? I thought a
> large sample size would always be a good thing.
>
> --
> Bülent Köksal

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/